classificationbox: saving model state

[laurenblackburn]

When I download a model state from classification box the prediction, examples, and classes are not saved. For example - I created a test model, model3, with 2 classes, 448 examples, and 8 predictions:

$ curl -X GET http://localhost:8084/classificationbox/models/model3/stats
{
    "success": true,
    "predictions": 8,
    "examples": 448,
    "classes": [
        {
            "name": "class1",
            "examples": 224
        },
        {
            "name": "class2",
            "examples": 224
        }
    ]
}

I download the model with

$ curl -X GET http://localhost:8084/classificationbox/state/model3 > model3.classificationbox

When I close machinebox and reopen it to test my saved state, model3's stats are altered and it appears that only the name and ID have been saved:

$ curl -X GET http://localhost:8084/classificationbox/models
{
    "success": true,
    "models": []
}
$ curl -X POST -F 'file=@model3.classificationbox' http://localhost:8084/classificationbox/state
{
    "success": true,
    "id": "model3",
    "name": "test",
    "options": {}
}
$ curl -XGET http://localhost:8084/classificationbox/models
{
    "success": true,
    "models": [
        {
            "id": "model3",
            "name": "test"
        }
    ]
}
$ curl -X GET http://localhost:8084/classificationbox/models/model3/stats
{
    "success": true,
    "predictions": 0,
    "examples": 0,
    "classes": []
}

How do I save the classes and examples for each model?

[dahernan]

The classes are probably saved correctly, try to do curl -XGET http://localhost:8084/classificationbox/models/model3 or run another prediction, to see that the classes are there

What are not saved with the model are the stats, the stats volatile right now.

Can you confirm this behavior?

[laurenblackburn]

Confirmed that the classes appear with I do curl -XGET http://localhost:8084/classificationbox/models/model3

Does this mean that there is no way to tell how many examples you've trained each class over if you train over more than one instance of classifcationbox?

[dahernan]

You need to count it on the client side right now, I don't think we are going to support saving the stats, because you can monitor that in many ways

[laurenblackburn]

Okay, thanks for the clarification

[dahernan]

No problem, let us know any other problem or question